It doesn't really know where this training data comes from. It probably isn't a problem yet, but at some point in the future it will be for sure. It already hap

It doesn't really know where this training data comes from. It probably isn't a problem yet, but at some point in the future it will be for sure. It already happened with many smaller languages in the world – the languages live their own life in "automatic translation engines are trained by data generated by translation engines" loop.